fix(usage): price new models without waiting a day for the rate table

[t3dotgg]

Claude Fable 5.1 usage showed $0.00 on the Usage page. The server fetches LiteLLM's rate table once every 24 hours, so a model that LiteLLM lists after the last fetch stays unpriced until the TTL runs out. Pressing refresh rescanned transcripts but kept the stale table. Claude Code's 1M context variant, written as `claude-fable-5-1[1m]`, never matched the table at all.

Two changes:

• The refresh button and pull-to-refresh now send `refreshRates` to every connected environment. Each server refetches the table inside the TTL, with a one minute floor so repeated presses do not hammer the source. Range changes and initial loads keep the daily cadence.
• Rate lookup strips a bracketed suffix such as `[1m]`, so the 1M tier prices at the base rate. That matches the existing rule of pricing at the base tier.

Covers web, desktop, and mobile. Tests cover the suffix lookup and the forced refetch inside the TTL. User docs updated.

Made with Claude Fable 5.1 in Claude Code.

🤖 Generated with Claude Code

---

Note

Medium Risk
Changes LiteLLM fetch timing and model rate matching for usage cost; bounded by TTL floor and single-flight locking, but refresh can increase outbound fetches when users hammer refresh.

Overview
Fixes $0.00 usage for newly listed models (e.g. Claude Fable 5.1) and for Claude Code’s [1m] context-tier model strings.

Server: Adds serverRefreshUsageRates RPC and UsageService.refreshRates, which refetches the LiteLLM table inside the normal 24h TTL (with a 1-minute floor and a semaphore so bursts share one HTTP fetch). Normal readSummary scans still use the daily cadence via ensureRates(false).

Clients (web + mobile): Usage refresh / pull-to-refresh now calls refreshUsageRates per environment first (failures ignored), then **refresh**es the usage summary so transcripts rescans pick up updated pricing.

Pricing lookup:stripVariantSuffix strips bracket suffixes (e.g. claude-fable-5-1[1m]) so rates match the base model name.

Tests and user usage docs updated.

^{Reviewed by Cursor Bugbot for commit d378d2e. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Add on-demand usage rate refresh so new models price without waiting a day

• Adds the server.refreshUsageRates WebSocket RPC and UsageService.refreshRates effect, which force a rate-table load bypassing the 24h TTL but respecting a new RATES_REFRESH_FLOOR_MS one-minute floor.
• Mobile and web useUsage hooks now trigger the rate-refresh RPC for each connected environment before rescanning usage summaries, and still rescan even when the pricing request fails.
• Concurrent rate loads are serialized through a single-permit semaphore in UsageService.make; client-side refresh commands use single-flight per environment.
• lookupRate now strips bracketed variant suffixes (e.g. context-tier) so a variant model resolves to its base model's rate.
• Risk: RATES_REFRESH_FLOOR_MS in UsageService.ts caps forced refreshes at once per minute; if the floor is too short, repeated refreshes could increase rate-table fetch load on the pricing source.

^{Macroscope summarized d378d2e.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.4 KiB	13.3 KiB	−167 B (−1.2%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	−5 B (−0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.6 KiB	6.4 KiB	−162 B (−2.4%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	57.0 KiB	55.6 KiB	−1.4 KiB (−2.5%)	66.4 KiB	✅
Codex	Live turn messages	10	10	0 (0.0%)	21	✅
Claude	Total thread wire	13.3 KiB	13.2 KiB	−84 B (−0.6%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	−10 B (−0.1%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.4 KiB	6.3 KiB	−74 B (−1.1%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	56.4 KiB	56.3 KiB	−88 B (−0.2%)	66.4 KiB	✅
Claude	Live turn messages	10	8	−2 (−20.0%)	21	✅

Baseline: 70cd258 · PR result: d378d2e · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 109.4 KiB
• Claude decoded thread snapshot: 110.1 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[cursor]

Cursor Bugbot has reviewed your changes using high effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 4a76454. Configure here.}

[macroscopeapp]

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This PR changes the usage refresh behavior, outbound rate-table fetching, and model pricing results across web, mobile, and server paths. It also modifies the server authorization mapping, which is a sensitive auth-directory change requiring human review.

^{You can add or adjust custom eligibility rules. Learn more.}

[t3dotgg]

Claude Fable 5.1 usage showed $0.00 on the Usage page. The server fetches LiteLLM's rate table once every 24 hours, so a model that LiteLLM lists after the last fetch stays unpriced until the TTL runs out. Pressing refresh rescanned transcripts but kept the stale table. Claude Code's 1M context variant, written as `claude-fable-5-1[1m]`, never matched the table at all.

Two changes:

• The refresh button and pull-to-refresh now send `refreshRates` to every connected environment. Each server refetches the table inside the TTL, with a one minute floor so repeated presses do not hammer the source. Range changes and initial loads keep the daily cadence.
• Rate lookup strips a bracketed suffix such as `[1m]`, so the 1M tier prices at the base rate. That matches the existing rule of pricing at the base tier.

Covers web, desktop, and mobile. Tests cover the suffix lookup and the forced refetch inside the TTL. User docs updated.

Made with Claude Fable 5.1 in Claude Code.

🤖 Generated with Claude Code

---

Note

Medium Risk
Changes LiteLLM fetch timing and model rate matching for usage cost; bounded by TTL floor and single-flight locking, but refresh can increase outbound fetches when users hammer refresh.

Overview
Fixes $0.00 usage for newly listed models (e.g. Claude Fable 5.1) and for Claude Code’s [1m] context-tier model strings.

Server: Adds serverRefreshUsageRates RPC and UsageService.refreshRates, which refetches the LiteLLM table inside the normal 24h TTL (with a 1-minute floor and a semaphore so bursts share one HTTP fetch). Normal readSummary scans still use the daily cadence via ensureRates(false).

Clients (web + mobile): Usage refresh / pull-to-refresh now calls refreshUsageRates per environment first (failures ignored), then **refresh**es the usage summary so transcripts rescans pick up updated pricing.

Pricing lookup:stripVariantSuffix strips bracket suffixes (e.g. claude-fable-5-1[1m]) so rates match the base model name.

Tests and user usage docs updated.

^{Reviewed by Cursor Bugbot for commit d378d2e. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Add on-demand usage rate refresh so new models price without waiting a day

• Adds the server.refreshUsageRates WebSocket RPC and UsageService.refreshRates effect, which force a rate-table load bypassing the 24h TTL but respecting a new RATES_REFRESH_FLOOR_MS one-minute floor.
• Mobile and web useUsage hooks now trigger the rate-refresh RPC for each connected environment before rescanning usage summaries, and still rescan even when the pricing request fails.
• Concurrent rate loads are serialized through a single-permit semaphore in UsageService.make; client-side refresh commands use single-flight per environment.
• lookupRate now strips bracketed variant suffixes (e.g. context-tier) so a variant model resolves to its base model's rate.
• Risk: RATES_REFRESH_FLOOR_MS in UsageService.ts caps forced refreshes at once per minute; if the floor is too short, repeated refreshes could increase rate-table fetch load on the pricing source.

^{Macroscope summarized d378d2e.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.4 KiB	13.3 KiB	−167 B (−1.2%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	−5 B (−0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.6 KiB	6.4 KiB	−162 B (−2.4%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	57.0 KiB	55.6 KiB	−1.4 KiB (−2.5%)	66.4 KiB	✅
Codex	Live turn messages	10	10	0 (0.0%)	21	✅
Claude	Total thread wire	13.3 KiB	13.2 KiB	−84 B (−0.6%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	−10 B (−0.1%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.4 KiB	6.3 KiB	−74 B (−1.1%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	56.4 KiB	56.3 KiB	−88 B (−0.2%)	66.4 KiB	✅
Claude	Live turn messages	10	8	−2 (−20.0%)	21	✅

Baseline: 70cd258 · PR result: d378d2e · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 109.4 KiB
• Claude decoded thread snapshot: 110.1 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[cursor]

Cursor Bugbot has reviewed your changes using high effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 4a76454. Configure here.}

[macroscopeapp]

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This PR changes the usage refresh behavior, outbound rate-table fetching, and model pricing results across web, mobile, and server paths. It also modifies the server authorization mapping, which is a sensitive auth-directory change requiring human review.

^{You can add or adjust custom eligibility rules. Learn more.}

[t3dotgg]

Claude Fable 5.1 usage showed $0.00 on the Usage page. The server fetches LiteLLM's rate table once every 24 hours, so a model that LiteLLM lists after the last fetch stays unpriced until the TTL runs out. Pressing refresh rescanned transcripts but kept the stale table. Claude Code's 1M context variant, written as `claude-fable-5-1[1m]`, never matched the table at all.

Two changes:

• The refresh button and pull-to-refresh now send `refreshRates` to every connected environment. Each server refetches the table inside the TTL, with a one minute floor so repeated presses do not hammer the source. Range changes and initial loads keep the daily cadence.
• Rate lookup strips a bracketed suffix such as `[1m]`, so the 1M tier prices at the base rate. That matches the existing rule of pricing at the base tier.

Covers web, desktop, and mobile. Tests cover the suffix lookup and the forced refetch inside the TTL. User docs updated.

Made with Claude Fable 5.1 in Claude Code.

🤖 Generated with Claude Code

---

Note

Medium Risk
Changes LiteLLM fetch timing and model rate matching for usage cost; bounded by TTL floor and single-flight locking, but refresh can increase outbound fetches when users hammer refresh.

Overview
Fixes $0.00 usage for newly listed models (e.g. Claude Fable 5.1) and for Claude Code’s [1m] context-tier model strings.

Server: Adds serverRefreshUsageRates RPC and UsageService.refreshRates, which refetches the LiteLLM table inside the normal 24h TTL (with a 1-minute floor and a semaphore so bursts share one HTTP fetch). Normal readSummary scans still use the daily cadence via ensureRates(false).

Clients (web + mobile): Usage refresh / pull-to-refresh now calls refreshUsageRates per environment first (failures ignored), then **refresh**es the usage summary so transcripts rescans pick up updated pricing.

Pricing lookup:stripVariantSuffix strips bracket suffixes (e.g. claude-fable-5-1[1m]) so rates match the base model name.

Tests and user usage docs updated.

^{Reviewed by Cursor Bugbot for commit d378d2e. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Add on-demand usage rate refresh so new models price without waiting a day

• Adds the server.refreshUsageRates WebSocket RPC and UsageService.refreshRates effect, which force a rate-table load bypassing the 24h TTL but respecting a new RATES_REFRESH_FLOOR_MS one-minute floor.
• Mobile and web useUsage hooks now trigger the rate-refresh RPC for each connected environment before rescanning usage summaries, and still rescan even when the pricing request fails.
• Concurrent rate loads are serialized through a single-permit semaphore in UsageService.make; client-side refresh commands use single-flight per environment.
• lookupRate now strips bracketed variant suffixes (e.g. context-tier) so a variant model resolves to its base model's rate.
• Risk: RATES_REFRESH_FLOOR_MS in UsageService.ts caps forced refreshes at once per minute; if the floor is too short, repeated refreshes could increase rate-table fetch load on the pricing source.

^{Macroscope summarized d378d2e.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.4 KiB	13.3 KiB	−167 B (−1.2%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	−5 B (−0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.6 KiB	6.4 KiB	−162 B (−2.4%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	57.0 KiB	55.6 KiB	−1.4 KiB (−2.5%)	66.4 KiB	✅
Codex	Live turn messages	10	10	0 (0.0%)	21	✅
Claude	Total thread wire	13.3 KiB	13.2 KiB	−84 B (−0.6%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	−10 B (−0.1%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.4 KiB	6.3 KiB	−74 B (−1.1%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	56.4 KiB	56.3 KiB	−88 B (−0.2%)	66.4 KiB	✅
Claude	Live turn messages	10	8	−2 (−20.0%)	21	✅

Baseline: 70cd258 · PR result: d378d2e · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 109.4 KiB
• Claude decoded thread snapshot: 110.1 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[cursor]

Cursor Bugbot has reviewed your changes using high effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 4a76454. Configure here.}

[macroscopeapp]

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This PR changes the usage refresh behavior, outbound rate-table fetching, and model pricing results across web, mobile, and server paths. It also modifies the server authorization mapping, which is a sensitive auth-directory change requiring human review.

^{You can add or adjust custom eligibility rules. Learn more.}

[t3dotgg]

Claude Fable 5.1 usage showed $0.00 on the Usage page. The server fetches LiteLLM's rate table once every 24 hours, so a model that LiteLLM lists after the last fetch stays unpriced until the TTL runs out. Pressing refresh rescanned transcripts but kept the stale table. Claude Code's 1M context variant, written as `claude-fable-5-1[1m]`, never matched the table at all.

Two changes:

• The refresh button and pull-to-refresh now send `refreshRates` to every connected environment. Each server refetches the table inside the TTL, with a one minute floor so repeated presses do not hammer the source. Range changes and initial loads keep the daily cadence.
• Rate lookup strips a bracketed suffix such as `[1m]`, so the 1M tier prices at the base rate. That matches the existing rule of pricing at the base tier.

Covers web, desktop, and mobile. Tests cover the suffix lookup and the forced refetch inside the TTL. User docs updated.

Made with Claude Fable 5.1 in Claude Code.

🤖 Generated with Claude Code

---

Note

Medium Risk
Changes LiteLLM fetch timing and model rate matching for usage cost; bounded by TTL floor and single-flight locking, but refresh can increase outbound fetches when users hammer refresh.

Overview
Fixes $0.00 usage for newly listed models (e.g. Claude Fable 5.1) and for Claude Code’s [1m] context-tier model strings.

Server: Adds serverRefreshUsageRates RPC and UsageService.refreshRates, which refetches the LiteLLM table inside the normal 24h TTL (with a 1-minute floor and a semaphore so bursts share one HTTP fetch). Normal readSummary scans still use the daily cadence via ensureRates(false).

Clients (web + mobile): Usage refresh / pull-to-refresh now calls refreshUsageRates per environment first (failures ignored), then **refresh**es the usage summary so transcripts rescans pick up updated pricing.

Pricing lookup:stripVariantSuffix strips bracket suffixes (e.g. claude-fable-5-1[1m]) so rates match the base model name.

Tests and user usage docs updated.

^{Reviewed by Cursor Bugbot for commit d378d2e. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Add on-demand usage rate refresh so new models price without waiting a day

• Adds the server.refreshUsageRates WebSocket RPC and UsageService.refreshRates effect, which force a rate-table load bypassing the 24h TTL but respecting a new RATES_REFRESH_FLOOR_MS one-minute floor.
• Mobile and web useUsage hooks now trigger the rate-refresh RPC for each connected environment before rescanning usage summaries, and still rescan even when the pricing request fails.
• Concurrent rate loads are serialized through a single-permit semaphore in UsageService.make; client-side refresh commands use single-flight per environment.
• lookupRate now strips bracketed variant suffixes (e.g. context-tier) so a variant model resolves to its base model's rate.
• Risk: RATES_REFRESH_FLOOR_MS in UsageService.ts caps forced refreshes at once per minute; if the floor is too short, repeated refreshes could increase rate-table fetch load on the pricing source.

^{Macroscope summarized d378d2e.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.4 KiB	13.3 KiB	−167 B (−1.2%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	−5 B (−0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.6 KiB	6.4 KiB	−162 B (−2.4%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	57.0 KiB	55.6 KiB	−1.4 KiB (−2.5%)	66.4 KiB	✅
Codex	Live turn messages	10	10	0 (0.0%)	21	✅
Claude	Total thread wire	13.3 KiB	13.2 KiB	−84 B (−0.6%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	−10 B (−0.1%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.4 KiB	6.3 KiB	−74 B (−1.1%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	56.4 KiB	56.3 KiB	−88 B (−0.2%)	66.4 KiB	✅
Claude	Live turn messages	10	8	−2 (−20.0%)	21	✅

Baseline: 70cd258 · PR result: d378d2e · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 109.4 KiB
• Claude decoded thread snapshot: 110.1 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[cursor]

Cursor Bugbot has reviewed your changes using high effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 4a76454. Configure here.}

[macroscopeapp]

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This PR changes the usage refresh behavior, outbound rate-table fetching, and model pricing results across web, mobile, and server paths. It also modifies the server authorization mapping, which is a sensitive auth-directory change requiring human review.

^{You can add or adjust custom eligibility rules. Learn more.}

[t3dotgg]

Claude Fable 5.1 usage showed $0.00 on the Usage page. The server fetches LiteLLM's rate table once every 24 hours, so a model that LiteLLM lists after the last fetch stays unpriced until the TTL runs out. Pressing refresh rescanned transcripts but kept the stale table. Claude Code's 1M context variant, written as `claude-fable-5-1[1m]`, never matched the table at all.

Two changes:

• The refresh button and pull-to-refresh now send `refreshRates` to every connected environment. Each server refetches the table inside the TTL, with a one minute floor so repeated presses do not hammer the source. Range changes and initial loads keep the daily cadence.
• Rate lookup strips a bracketed suffix such as `[1m]`, so the 1M tier prices at the base rate. That matches the existing rule of pricing at the base tier.

Covers web, desktop, and mobile. Tests cover the suffix lookup and the forced refetch inside the TTL. User docs updated.

Made with Claude Fable 5.1 in Claude Code.

🤖 Generated with Claude Code

---

Note

Medium Risk
Changes LiteLLM fetch timing and model rate matching for usage cost; bounded by TTL floor and single-flight locking, but refresh can increase outbound fetches when users hammer refresh.

Overview
Fixes $0.00 usage for newly listed models (e.g. Claude Fable 5.1) and for Claude Code’s [1m] context-tier model strings.

Server: Adds serverRefreshUsageRates RPC and UsageService.refreshRates, which refetches the LiteLLM table inside the normal 24h TTL (with a 1-minute floor and a semaphore so bursts share one HTTP fetch). Normal readSummary scans still use the daily cadence via ensureRates(false).

Clients (web + mobile): Usage refresh / pull-to-refresh now calls refreshUsageRates per environment first (failures ignored), then **refresh**es the usage summary so transcripts rescans pick up updated pricing.

Pricing lookup:stripVariantSuffix strips bracket suffixes (e.g. claude-fable-5-1[1m]) so rates match the base model name.

Tests and user usage docs updated.

^{Reviewed by Cursor Bugbot for commit d378d2e. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Add on-demand usage rate refresh so new models price without waiting a day

• Adds the server.refreshUsageRates WebSocket RPC and UsageService.refreshRates effect, which force a rate-table load bypassing the 24h TTL but respecting a new RATES_REFRESH_FLOOR_MS one-minute floor.
• Mobile and web useUsage hooks now trigger the rate-refresh RPC for each connected environment before rescanning usage summaries, and still rescan even when the pricing request fails.
• Concurrent rate loads are serialized through a single-permit semaphore in UsageService.make; client-side refresh commands use single-flight per environment.
• lookupRate now strips bracketed variant suffixes (e.g. context-tier) so a variant model resolves to its base model's rate.
• Risk: RATES_REFRESH_FLOOR_MS in UsageService.ts caps forced refreshes at once per minute; if the floor is too short, repeated refreshes could increase rate-table fetch load on the pricing source.

^{Macroscope summarized d378d2e.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.4 KiB	13.3 KiB	−167 B (−1.2%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	−5 B (−0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.6 KiB	6.4 KiB	−162 B (−2.4%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	57.0 KiB	55.6 KiB	−1.4 KiB (−2.5%)	66.4 KiB	✅
Codex	Live turn messages	10	10	0 (0.0%)	21	✅
Claude	Total thread wire	13.3 KiB	13.2 KiB	−84 B (−0.6%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	−10 B (−0.1%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.4 KiB	6.3 KiB	−74 B (−1.1%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	56.4 KiB	56.3 KiB	−88 B (−0.2%)	66.4 KiB	✅
Claude	Live turn messages	10	8	−2 (−20.0%)	21	✅

Baseline: 70cd258 · PR result: d378d2e · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 109.4 KiB
• Claude decoded thread snapshot: 110.1 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[cursor]

Cursor Bugbot has reviewed your changes using high effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 4a76454. Configure here.}

[macroscopeapp]

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This PR changes the usage refresh behavior, outbound rate-table fetching, and model pricing results across web, mobile, and server paths. It also modifies the server authorization mapping, which is a sensitive auth-directory change requiring human review.

^{You can add or adjust custom eligibility rules. Learn more.}

[t3dotgg]

Claude Fable 5.1 usage showed $0.00 on the Usage page. The server fetches LiteLLM's rate table once every 24 hours, so a model that LiteLLM lists after the last fetch stays unpriced until the TTL runs out. Pressing refresh rescanned transcripts but kept the stale table. Claude Code's 1M context variant, written as `claude-fable-5-1[1m]`, never matched the table at all.

Two changes:

• The refresh button and pull-to-refresh now send `refreshRates` to every connected environment. Each server refetches the table inside the TTL, with a one minute floor so repeated presses do not hammer the source. Range changes and initial loads keep the daily cadence.
• Rate lookup strips a bracketed suffix such as `[1m]`, so the 1M tier prices at the base rate. That matches the existing rule of pricing at the base tier.

Covers web, desktop, and mobile. Tests cover the suffix lookup and the forced refetch inside the TTL. User docs updated.

Made with Claude Fable 5.1 in Claude Code.

🤖 Generated with Claude Code

---

Note

Medium Risk
Changes LiteLLM fetch timing and model rate matching for usage cost; bounded by TTL floor and single-flight locking, but refresh can increase outbound fetches when users hammer refresh.

Overview
Fixes $0.00 usage for newly listed models (e.g. Claude Fable 5.1) and for Claude Code’s [1m] context-tier model strings.

Server: Adds serverRefreshUsageRates RPC and UsageService.refreshRates, which refetches the LiteLLM table inside the normal 24h TTL (with a 1-minute floor and a semaphore so bursts share one HTTP fetch). Normal readSummary scans still use the daily cadence via ensureRates(false).

Clients (web + mobile): Usage refresh / pull-to-refresh now calls refreshUsageRates per environment first (failures ignored), then **refresh**es the usage summary so transcripts rescans pick up updated pricing.

Pricing lookup:stripVariantSuffix strips bracket suffixes (e.g. claude-fable-5-1[1m]) so rates match the base model name.

Tests and user usage docs updated.

^{Reviewed by Cursor Bugbot for commit d378d2e. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Add on-demand usage rate refresh so new models price without waiting a day

• Adds the server.refreshUsageRates WebSocket RPC and UsageService.refreshRates effect, which force a rate-table load bypassing the 24h TTL but respecting a new RATES_REFRESH_FLOOR_MS one-minute floor.
• Mobile and web useUsage hooks now trigger the rate-refresh RPC for each connected environment before rescanning usage summaries, and still rescan even when the pricing request fails.
• Concurrent rate loads are serialized through a single-permit semaphore in UsageService.make; client-side refresh commands use single-flight per environment.
• lookupRate now strips bracketed variant suffixes (e.g. context-tier) so a variant model resolves to its base model's rate.
• Risk: RATES_REFRESH_FLOOR_MS in UsageService.ts caps forced refreshes at once per minute; if the floor is too short, repeated refreshes could increase rate-table fetch load on the pricing source.

^{Macroscope summarized d378d2e.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.4 KiB	13.3 KiB	−167 B (−1.2%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	−5 B (−0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.6 KiB	6.4 KiB	−162 B (−2.4%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	57.0 KiB	55.6 KiB	−1.4 KiB (−2.5%)	66.4 KiB	✅
Codex	Live turn messages	10	10	0 (0.0%)	21	✅
Claude	Total thread wire	13.3 KiB	13.2 KiB	−84 B (−0.6%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	−10 B (−0.1%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.4 KiB	6.3 KiB	−74 B (−1.1%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	56.4 KiB	56.3 KiB	−88 B (−0.2%)	66.4 KiB	✅
Claude	Live turn messages	10	8	−2 (−20.0%)	21	✅

Baseline: 70cd258 · PR result: d378d2e · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 109.4 KiB
• Claude decoded thread snapshot: 110.1 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[cursor]

Cursor Bugbot has reviewed your changes using high effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 4a76454. Configure here.}

[macroscopeapp]

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This PR changes the usage refresh behavior, outbound rate-table fetching, and model pricing results across web, mobile, and server paths. It also modifies the server authorization mapping, which is a sensitive auth-directory change requiring human review.

^{You can add or adjust custom eligibility rules. Learn more.}

[t3dotgg]

Claude Fable 5.1 usage showed $0.00 on the Usage page. The server fetches LiteLLM's rate table once every 24 hours, so a model that LiteLLM lists after the last fetch stays unpriced until the TTL runs out. Pressing refresh rescanned transcripts but kept the stale table. Claude Code's 1M context variant, written as `claude-fable-5-1[1m]`, never matched the table at all.

Two changes:

• The refresh button and pull-to-refresh now send `refreshRates` to every connected environment. Each server refetches the table inside the TTL, with a one minute floor so repeated presses do not hammer the source. Range changes and initial loads keep the daily cadence.
• Rate lookup strips a bracketed suffix such as `[1m]`, so the 1M tier prices at the base rate. That matches the existing rule of pricing at the base tier.

Covers web, desktop, and mobile. Tests cover the suffix lookup and the forced refetch inside the TTL. User docs updated.

Made with Claude Fable 5.1 in Claude Code.

🤖 Generated with Claude Code

---

Note

Medium Risk
Changes LiteLLM fetch timing and model rate matching for usage cost; bounded by TTL floor and single-flight locking, but refresh can increase outbound fetches when users hammer refresh.

Overview
Fixes $0.00 usage for newly listed models (e.g. Claude Fable 5.1) and for Claude Code’s [1m] context-tier model strings.

Server: Adds serverRefreshUsageRates RPC and UsageService.refreshRates, which refetches the LiteLLM table inside the normal 24h TTL (with a 1-minute floor and a semaphore so bursts share one HTTP fetch). Normal readSummary scans still use the daily cadence via ensureRates(false).

Clients (web + mobile): Usage refresh / pull-to-refresh now calls refreshUsageRates per environment first (failures ignored), then **refresh**es the usage summary so transcripts rescans pick up updated pricing.

Pricing lookup:stripVariantSuffix strips bracket suffixes (e.g. claude-fable-5-1[1m]) so rates match the base model name.

Tests and user usage docs updated.

^{Reviewed by Cursor Bugbot for commit d378d2e. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Add on-demand usage rate refresh so new models price without waiting a day

• Adds the server.refreshUsageRates WebSocket RPC and UsageService.refreshRates effect, which force a rate-table load bypassing the 24h TTL but respecting a new RATES_REFRESH_FLOOR_MS one-minute floor.
• Mobile and web useUsage hooks now trigger the rate-refresh RPC for each connected environment before rescanning usage summaries, and still rescan even when the pricing request fails.
• Concurrent rate loads are serialized through a single-permit semaphore in UsageService.make; client-side refresh commands use single-flight per environment.
• lookupRate now strips bracketed variant suffixes (e.g. context-tier) so a variant model resolves to its base model's rate.
• Risk: RATES_REFRESH_FLOOR_MS in UsageService.ts caps forced refreshes at once per minute; if the floor is too short, repeated refreshes could increase rate-table fetch load on the pricing source.

^{Macroscope summarized d378d2e.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.4 KiB	13.3 KiB	−167 B (−1.2%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	−5 B (−0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.6 KiB	6.4 KiB	−162 B (−2.4%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	57.0 KiB	55.6 KiB	−1.4 KiB (−2.5%)	66.4 KiB	✅
Codex	Live turn messages	10	10	0 (0.0%)	21	✅
Claude	Total thread wire	13.3 KiB	13.2 KiB	−84 B (−0.6%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	−10 B (−0.1%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.4 KiB	6.3 KiB	−74 B (−1.1%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	56.4 KiB	56.3 KiB	−88 B (−0.2%)	66.4 KiB	✅
Claude	Live turn messages	10	8	−2 (−20.0%)	21	✅

Baseline: 70cd258 · PR result: d378d2e · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 109.4 KiB
• Claude decoded thread snapshot: 110.1 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[cursor]

Cursor Bugbot has reviewed your changes using high effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 4a76454. Configure here.}

[macroscopeapp]

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This PR changes the usage refresh behavior, outbound rate-table fetching, and model pricing results across web, mobile, and server paths. It also modifies the server authorization mapping, which is a sensitive auth-directory change requiring human review.

^{You can add or adjust custom eligibility rules. Learn more.}

[t3dotgg]

Claude Fable 5.1 usage showed $0.00 on the Usage page. The server fetches LiteLLM's rate table once every 24 hours, so a model that LiteLLM lists after the last fetch stays unpriced until the TTL runs out. Pressing refresh rescanned transcripts but kept the stale table. Claude Code's 1M context variant, written as `claude-fable-5-1[1m]`, never matched the table at all.

Two changes:

• The refresh button and pull-to-refresh now send `refreshRates` to every connected environment. Each server refetches the table inside the TTL, with a one minute floor so repeated presses do not hammer the source. Range changes and initial loads keep the daily cadence.
• Rate lookup strips a bracketed suffix such as `[1m]`, so the 1M tier prices at the base rate. That matches the existing rule of pricing at the base tier.

Covers web, desktop, and mobile. Tests cover the suffix lookup and the forced refetch inside the TTL. User docs updated.

Made with Claude Fable 5.1 in Claude Code.

🤖 Generated with Claude Code

---

Note

Medium Risk
Changes LiteLLM fetch timing and model rate matching for usage cost; bounded by TTL floor and single-flight locking, but refresh can increase outbound fetches when users hammer refresh.

Overview
Fixes $0.00 usage for newly listed models (e.g. Claude Fable 5.1) and for Claude Code’s [1m] context-tier model strings.

Server: Adds serverRefreshUsageRates RPC and UsageService.refreshRates, which refetches the LiteLLM table inside the normal 24h TTL (with a 1-minute floor and a semaphore so bursts share one HTTP fetch). Normal readSummary scans still use the daily cadence via ensureRates(false).

Clients (web + mobile): Usage refresh / pull-to-refresh now calls refreshUsageRates per environment first (failures ignored), then **refresh**es the usage summary so transcripts rescans pick up updated pricing.

Pricing lookup:stripVariantSuffix strips bracket suffixes (e.g. claude-fable-5-1[1m]) so rates match the base model name.

Tests and user usage docs updated.

^{Reviewed by Cursor Bugbot for commit d378d2e. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Add on-demand usage rate refresh so new models price without waiting a day

• Adds the server.refreshUsageRates WebSocket RPC and UsageService.refreshRates effect, which force a rate-table load bypassing the 24h TTL but respecting a new RATES_REFRESH_FLOOR_MS one-minute floor.
• Mobile and web useUsage hooks now trigger the rate-refresh RPC for each connected environment before rescanning usage summaries, and still rescan even when the pricing request fails.
• Concurrent rate loads are serialized through a single-permit semaphore in UsageService.make; client-side refresh commands use single-flight per environment.
• lookupRate now strips bracketed variant suffixes (e.g. context-tier) so a variant model resolves to its base model's rate.
• Risk: RATES_REFRESH_FLOOR_MS in UsageService.ts caps forced refreshes at once per minute; if the floor is too short, repeated refreshes could increase rate-table fetch load on the pricing source.

^{Macroscope summarized d378d2e.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.4 KiB	13.3 KiB	−167 B (−1.2%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	−5 B (−0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.6 KiB	6.4 KiB	−162 B (−2.4%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	57.0 KiB	55.6 KiB	−1.4 KiB (−2.5%)	66.4 KiB	✅
Codex	Live turn messages	10	10	0 (0.0%)	21	✅
Claude	Total thread wire	13.3 KiB	13.2 KiB	−84 B (−0.6%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	−10 B (−0.1%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.4 KiB	6.3 KiB	−74 B (−1.1%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	56.4 KiB	56.3 KiB	−88 B (−0.2%)	66.4 KiB	✅
Claude	Live turn messages	10	8	−2 (−20.0%)	21	✅

Baseline: 70cd258 · PR result: d378d2e · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 109.4 KiB
• Claude decoded thread snapshot: 110.1 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[cursor]

Cursor Bugbot has reviewed your changes using high effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 4a76454. Configure here.}

[macroscopeapp]

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This PR changes the usage refresh behavior, outbound rate-table fetching, and model pricing results across web, mobile, and server paths. It also modifies the server authorization mapping, which is a sensitive auth-directory change requiring human review.

^{You can add or adjust custom eligibility rules. Learn more.}